In the blog post Chaosbased encryption I introduced
a chaotic cryptology algorithm^{1}
using a Poincaré successor ordinal
function. In this article I want to present you a command line suite providing this very
Chaosbased encryption functionality alongside some extensions
as well as comparison to the AES algorithm.
This chaosbased cryptography suite is intended for educational purposes only:
As I am not a mathematician, I neither can guarantee the algorithm nor the program
code implementing the algorithm to be water proof^{2} or free of bugs! Any usage
is intended for educational purposes only and any other usage is at your own risk.
The show case
To evaluate the Chaosbased encryption command line
suite I prepared some sample images A, B,
C and D being similar to each
other but different in some obvious details (except the last image being a photograph).
Having those similar but differing images we can analyze the images (after being
encrypted with a given algorithm) by comparing them with each other (samples)
to get an idea on the effectiveness of the according algorithm.
Given similar but differing images we can analyze the accordingly encrypted images
by visual comparison to get an impression of the effectiveness of the according
algorithm.
As I already mentioned I am not mathematician, hence I need some way to estimate
the randomness and effectiveness of this command line suite by digging neither too
deep into cryptography mathematics nor into abstract algebra. For a start I therefore
decided to do some (non comprehensive) visual analysis (samples).
The setup
Given images A, B, C
and D, I use a toolchain to extract the raw pixel data
of each image, encrypt the raw pixel data with a given algorithm and convert the
result back (interpreted as raw pixel data) into an image of the same dimensions.
In fact I used the PNG
file format though the file format actually does not play a role. I could have
chosen any other file format (supporting the required lossless compression
and color depth ) and the visual result would have been the same as the analysis
focuses on the raw pixel data (using PICDAT
)
and not on the physical image file as a whole (with its file header, compression
method and so on).
The toolchain for the various steps during analysis consist of these four tools
(the first two being available in the downloads section of this site:
PICDAT
: Extracts raw pixel data from an image and the other way round
CHAOS
: Encrypts and decrypts data based on Chaosbased encryption
OPENSSL
: Encrypts and decrypts data based (amongst others) on the AES algorithm
ENT
: Pseudorandom number sequence test for calculating the entropy on a stream of bytes
COMPARE
: Mathematically and visually annotate the difference between an image and its reconstruction
The tool PICDAT
I dedicatedly developed
to accomplish visually analysing of (amongst others) Chaosbased encryption.
Please see the sources for the
actual compare*.sh
as well as validate*.sh
scripts automating the visual
analysis as described below! The DADUM
tool is similar in that
I have developed it for creating loads of data to test your algorithms (though
it was not needed in this setup).
Overview
First of all, the raw image data of the sample images A,
B, C and D
is extracted using PICDAT
^{3}. Then the raw
image data of each image is being encrypted with various flavors of the AES
and
the CHAOS
algorithms respectively (see below), using openssl
for the AES*
variants and CHAOS
^{3} for
the CHAOS*
variants.

A 
B 
C 
D 
Original 




AES256ECB 




AES256ECBSALTED 




AES256CBC 




AES256CBCSALTED 




CHAOSNONE 




CHAOSXOR 




CHAOSMUTATE 




CHAOSPREFIX32 




CHAOSM.XOR 




CHAOSM.P.32 




CHAOSM.XORP.32 




CHAOSSALTED 




CHAOSCERT 




Taking a look at image A (raw) data’s metrics with no encryption
applied we get the following values:
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Entropy bits/byte (8 = best): 0
 Arithmetic mean (127.5 = random): 255
 Serial correlation (none = 0.0): 100000
 Max compression (0% = none) = 100%
As the image is a blank (white) image, its raw data consists of an accordingly
large number of bytes with values 255
(0xFF
) as a white pixel is represented
by a sequence of bytes with (again) a value of 255
(0xFF
). Hence we get an
arithmetic mean of 255
(0xFF
).
For a good cryptographic algorithm I would expect the entropy (bits per byte)
to be very near to 8, the arithmetic mean to be very near to 127.5,
the serial correlation to be very near to 0.0 and the max compression
to be very near to 0%.
Observations
The below “Samples” section illustrates the various algorithms’ variants
by comparing two very similar encrypted images A, B
and C with each other: The bottom right images each are
the result of a COMPARE
operation
between the according (encrypted) image directly to the left and the according
(encrypted) image directly above that resulting image: Same areas between both
images are colored gray ■ in the resulting image,
differing areas are colored red ■, so that the
difference between two images may give a hint on the content of the images (a fully
red area as a result of comparison would be aspirable).
AES
At a first sight, AES256 (ECB) encryption still provides a quite
good impression on what has been encrypted, even AES256 (ECBSALTED)
salting^{4} does not ease the situation. This is a known “feature” of of the
electronic codebook
(ECB) mode used here. Furthermore, the comparison of the AES256 (ECB)
encrypted images allows conclusions to be drawn about the original images’ content
(see the according bottom right images). At least salting^{4} with AES256 (ECBSALTED)
makes comparing two very similar encrypted images challenging (notice the according
bottom right image being fully red, indicating no equal areas). Having a glance
at the encrypted images we are better off with AES256 (CBC), providing
no immediate hint on the content. Still, comparing two very similar images encrypted
using AES256 (CBC) reveal similar areas till the position where
the images actually differ the fist time (notice the according bottom right image’s
gray areas, indicating equal areas). Using AES256 (CBCSALTED)
looks perfect in that the comparison reveals no equal areas (notice the according
bottom right image being fully red, indicating no equal areas) and the encrypted
images look fully random.
CHAOS
As of the CHAOS comparisons we can see the evolution of the CHAOS
command: Vanilla plain CHAOS (NONE) encryption does not reveal immediate
correlation to the content of the original images (there seems already to be a quite
good randomness), although the comparison of the CHAOS (NONE) encrypted
images gives a first impression of the original images’ content (see the according
bottom right images). To soothe this situation, CHAOS (XOR), CHAOS (PREFIX32)
as well as CHAOS (MUTATE) flavors adding some obfuscation were
introduced without a convincing result. At least the combination of all three CHAOS (MUTATEXORPREFIX32)
variants seems to do the trick, any other combination “did not work out”. To not
rely on this combination of obfuscations, as this could have been pure coincidence
(regarding the chosen images), a salting CHAOS (SALTED) flavor
has been implemented. This reliably prevents easily getting an idea of the content
by simply comparing the encrypted images. For scaling purposes, the CHAOS (CERT)
addition has been added, providing means to apply an arbitrary number of encryption
passes using a certificate of detailed instructions for each pass: Hence multiple
passes with a mixture of obfuscation methods can be applied, including XOR
, PREFIX
,
MUTATE
or SLATED
obfuscations, alongside individual encryption keys, elongating
the resulting “key” and hardening against brute force attacks.
Critical reflection
In their publication on ChaosBased Cryptography: End of the Road?, Iercan, D., Dranga, O., Dragan, F. and Banias, O identify a weakness of chaosbased encryption being the “dynamic degradation of digital chaotic systems”:
“… Chaosbased cryptography emerged in the early 1990s as an innovative application of nonlinear dynamics in the chaotic regime. Even if in theory chaotic dynamics was thought to evolve into a new revolution in cryptography, in reallife an efficient and reliable chaosbased cryptosystem didn’t emerge. The main but not the only reason is the dynamic degradation of digital chaotic systems, a subject that became very popular in the last few years. This paper presents a new theoretical background related to this issue that proves the inefficiency of chaosbased encryption algorithms. Even more, in one of the two relevant case studies presented, another myth is demolished: the analog encryption base on synchronized chaos …” ChaosBased Cryptography: End of the Road?
(I cannot tell whether this applies to Chaosbased encryption using a Poincaré successor ordinal function as described in this article, or not)
Samples
The below samples illustrate the various algorithms’ variants by comparing two
very similar encrypted images A, B
and C with each other: The bottom right images each are
the result of a COMPARE
operation
between the according (encrypted) image directly to the left and the according (encrypted)
image directly above that resulting image: Same areas between both images are colored
gray ■ in the resulting image, differing areas
are colored red ■, so that the difference between
two images may give a hint on the content of the images (a fully red area as a result
of comparison would be aspirable).
AES
AES256 (ECB) encryption
The encryption still provides quite good impression on what has been encrypted.
This is a known “feature” of of the electronic codebook
(ECB) mode used here. Furthermore, the comparison of the encrypted images allows
conclusions to be drawn about the original images’ content (see the according bottom
right images).
Facts
 Algorithm: AES256ECB
 Key size: 256 bits (32 Bytes)
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 16^{5}
 Entropy bits/byte (8 = best): 4.000186
 Arithmetic mean (127.5 = random): 102.062991
 Serial correlation (none = 0.0): 0.248134
 Max compression (0% = none): 49%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



AES256 (ECBSALTED) encryption
Encryption still provides a quite good impression on what has been encrypted, although
salting^{4} has been applied. This is a known “feature” of of the
electronic codebook
(ECB) mode used here. At least salting^{4} makes comparing two very
similar encrypted images challenging (notice the according bottom right image being
fully red, indicating no equal areas).
Facts
 Algorithm: AES256ECBSALTED^{4}
 Key size: 256 bits (32 Bytes)
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 16^{5}
 Entropy bits/byte (8 = best): 4.000373
 Arithmetic mean (127.5 = random): 90.938478
 Serial correlation (none = 0.0): 0.065746
 Max compression (0% = none): 49%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



AES256 (CBC) encryption
Encryption does not provide an immediate hint on the content. Still, comparing two
very similar images reveal similar areas till the position where the images actually
differ the fist time (notice the according bottom right image’s gray areas, indicating
equal areas).
Facts
 Algorithm: AES256CBC
 Key size: 256 bits (32 Bytes)
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 16^{5}
 Entropy bits/byte (8 = best): 7.999896
 Arithmetic mean (127.5 = random): 127.455223
 Serial correlation (none = 0.0): 0.000401
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



AES256 (CBCSALTED) encryption
Salted ^{4} encryption looks perfect in that the comparison reveals no equal
areas (notice the according bottom right image being fully red, indicating no equal
areas) and the encrypted images look fully random.
Facts
 Algorithm: AES256CBCSALTED^{4}
 Key size: 256 bits (32 Bytes)
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 16^{5}
 Entropy bits/byte (8 = best): 7.999867
 Arithmetic mean (127.5 = random): 127.536467
 Serial correlation (none = 0.0): 0.000888
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



CHAOS
CHAOS (NONE) encryption
Vanilla plain encryption does not reveal immediate correlation to the content of
the original images (there seems already to be a quite good randomness), although
the comparison of the encrypted images gives a first impression of the original
images’ content (see the according bottom right images).
Facts
 Algorithm: CHAOSNONE
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 0
 Entropy bits/byte (8 = best): 7.999869
 Arithmetic mean (127.5 = random): 127.420840
 Serial correlation (none = 0.0): 0.000196
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



CHAOS (XOR) encryption
Adding some XOR
obfuscation during encryption does not lead to a convincing result.
Comparing two very similar images not only reveal similar areas till the position
where the images actually differ the fist time (notice the according bottom right
image’s gray areas, indicating equal areas).
Facts
 Algorithm: CHAOSXOR
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 0
 Entropy bits/byte (8 = best): 7.999878
 Arithmetic mean (127.5 = random): 127.535873
 Serial correlation (none = 0.0): 0.002190
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



CHAOS (PREFIX32) encryption
Adding some PREFIX
obfuscation during encryption does not lead to any improvement
at all. The comparison of the encrypted images gives the same impression of the
original images’ content as if no obfuscation at all would have been applied (see
the according bottom right images). Actually, PREFIX
obfuscation is not an obfuscation
mode by itself, it only shows an effect when hardening the XOR
and/or the MUTATE
obfuscation!
Facts
 Algorithm: CHAOSPREFIX32
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 32
 Entropy bits/byte (8 = best): 7.999855
 Arithmetic mean (127.5 = random): 127.461611
 Serial correlation (none = 0.0): 0.000197
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



CHAOS (MUTATE) encryption
Adding some MUTATE
obfuscation during encryption does not lead to a convincing
result. Comparing two very similar images not only reveal similar areas till the position
where the images actually differ the fist time (notice the according bottom right
image’s gray areas, indicating equal areas).
Facts
 Algorithm: CHAOSMUTATE
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 0
 Entropy bits/byte (8 = best): 7.999867
 Arithmetic mean (127.5 = random): 127.506294
 Serial correlation (none = 0.0): 0.000554
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



CHAOS (MUTATEXOR) encryption
Combining MUTATE
obfuscation with XOR
mutation during encryption does not
lead to a convincing result. Comparing two very similar images reveal similar areas
till the position where the images actually differ the fist time (notice the according
bottom right image’s gray areas, indicating equal areas).
Facts
 Algorithm: CHAOSMUTATEXOR
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 0
 Entropy bits/byte (8 = best): 7.999898
 Arithmetic mean (127.5 = random): 127.500697
 Serial correlation (none = 0.0): 0.000449
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



CHAOS (MUTATEPREFIX32) encryption
Combining MUTATE
obfuscation with PREFIX
mutation during encryption does not
lead to a convincing result. Comparing two very similar images still reveal few
similar areas (notice the according bottom right image’s gray areas, indicating
equal areas).
Facts
 Algorithm: CHAOSMUTATEPREFIX32
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 32
 Entropy bits/byte (8 = best): 7.999874
 Arithmetic mean (127.5 = random): 127.545626
 Serial correlation (none = 0.0): 0.001285
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



CHAOS (MUTATEXORPREFIX32) encryption
The combination of all three MUTATE
+ XOR
+ PREFIX32
obfuscation variants
makes comparing two very similar encrypted images hard (notice the according bottom
right image being fully red, indicating no equal areas).
Facts
 Algorithm: CHAOSMUTATEXORPREFIX32
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 32
 Entropy bits/byte (8 = best): 7.999880
 Arithmetic mean (127.5 = random): 127.527029
 Serial correlation (none = 0.0): 0.000805
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



CHAOS (SALTED) encryption
Salted encryption looks perfect in that the comparison reveals no equal areas (notice
the according bottom right image being fully red, indicating no equal areas) and
the encrypted images look fully random.
Facts
 Algorithm: CHAOSSALTED
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 25
 Entropy bits/byte (8 = best): 7.999869
 Arithmetic mean (127.5 = random): 127.530544
 Serial correlation (none = 0.0): 0.000776
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



CHAOS (CERT) encryption
For scaling purposes, means are provided to apply an arbitrary number of encryption
passes using a certificate of detailed instructions for each pass: Hence multiple
passes with a mixture of obfuscation methods are applied, including XOR
, PREFIX
,
MUTATE
or SLATED
obfuscations, alongside individual encryption keys, elongating
the resulting “key” and hardening against brute force attacks.
Facts
 Algorithm: CHAOSCERT
 Chain length: 8
 First key: Salted = true, XOR = false, Mutate = false, Prefix = none
 Second key: Salted = false, XOR = false, Mutate = true, Prefix = none
 Other keys: Salted = false, XOR = false, Mutate = false, Prefix = none
 Image file: A (blank image)
 Image width: 800px
 Image height: 600px
 Algorithm prefix (bytes): 25
 Entropy bits/byte (8 = best): 7.999887
 Arithmetic mean (127.5 = random): 127.501808
 Serial correlation (none = 0.0): 0.000777
 Max compression (0% = none): 0%
A + B

Original 
Encrypted 
B 
Original 



Encrypted 

enc(A) ≏ enc(B) ~ (A,B) 

A 



B + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(B) ≏ enc(C) ~ (B,C) 

B 



A + C

Original 
Encrypted 
C 
Original 



Encrypted 

enc(A) ≏ enc(C) ~ (A,C) 

A 



See also